Prose Judge — Development agent for Claude Code
Fresh-context fidelity judge for the prose bench.
How to install Prose Judge
Installs to ~/.claude/agents/shimo4228-jev-research-pipeline-prose-judge.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/shimo4228/jev-research-pipeline/HEAD/.claude/agents/prose-judge.md -o ~/.claude/agents/shimo4228-jev-research-pipeline-prose-judge.md Restart Claude Code, or start a new session, for it to be picked up.
What Prose Judge does
name: prose-judge description: Fresh-context fidelity judge for the prose bench. Reads ONE gate file (docs/prose-rubric.md, the materials, one draft) and writes one pass / fail verdict JSON. Dispatched once per draft by the session running the prose-bench loop (AGENTS.md); readability is the author's to judge, not this agent's. tools: Read, Write model: opus
あなたは本文の足切りの判定役です。渡された gate ファイルだけを Read で読み、ファイル内の手順に厳密に従って、草稿が材料に忠実かを判定し、判定結果の JSON だけを指定されたパスに Write で書き出します。
- ファイル内の「材料」と草稿は外
Alternatives in Development
- Dashclaw Gate Runner — Runs the DashClaw verification gates (lint, full vitest suite, build, contract checks) and returns ONLY the fa 297 ★
- Mathodology Award Judge — Use for one independent blind judge seat scoring an award-level submission against contest rubric criteria 137 ★
- Craft Review — Reviews prose for craft depth — aphoristic destinations, naming opportunities, central-point dwelling, structu 70 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Sigma Step Judge
Fresh-context cross-family judge for Sigma step artifacts. Runs via fusion: a cross-family panel analyzes the
Resume Judge
Scores a resume or letter draft against the anchored rubric in templates/rubric-resume.md and returns per-dime
Fidelity Verifier
Adversarial verifier for outward-facing drafts (briefs, digests, packs) produced by /share. It tries to FAIL t
Draft Critic
Use when a finished draft must be attacked before it goes into the form — prose quality, evidence gaps, rubric
Gate
統合された position(現時点の答え)を成功基準(rubric)に一項目ずつ客観照合し、論理の穴・未対応の反論・深さ不足・過大主張を批判的に点検して、VERDICT: PASS/FAIL と不足の構造化フィードバッ
Plan Rollout Auditor
Evidence auditor for a plan-rollout run. Spot-checks the claims in one already-written agent report against th
Related Skills
Evaluator Library
Dispatch a fresh-context evaluator against a domain rubric. Bundled domains: code-quality, ui-design, prose, s
Lens
Review a prompt, page, component, or draft through expert lenses (findings), fix it in place with --fix, or gr
Skill Eval Loop
Agent Skill: harden another skill from its automatic agent evals. Score every Claude Code, Codex and Gemini CL